Professional Documents
Culture Documents
Accurate prediction of phenological development in maize (Zea mays L.) is fundamental to determining crop adaptation and yield
potential. A number of thermal functions are used in crop models, but their relative precision in predicting maize development has
not been quantified. The objectives of this study were (i) to evaluate the precision of eight thermal functions, (ii) to assess the effects
of source data on the ability to differentiate among thermal functions, and (iii) to attribute the precision of thermal functions to their
response across various temperature ranges. Data sets used in this study represent >1000 distinct maize hybrids, >50 geographic
locations, and multiple planting dates and years. Thermal functions and calendar days were evaluated and grouped based on their
temperature response and derivation as empirical linear, empirical nonlinear, and process-based functions. Precision in predicting
phase durations from planting to anthesis or silking and from silking to physiological maturity was evaluated. Large data sets
enabled increased differentiation of thermal functions, even when smaller data sets contained orthogonal, multi-location and -year
data. At the highest level of differentiation, precision of thermal functions was in the order calendar days < empirical linear < process
based < empirical nonlinear. Precision was associated with relatively low temperature sensitivity across the 10 to 26C range. In
contrast to other thermal functions, process-based functions were derived using supra-optimal temperatures, and consequently,
they may better represent the developmental response of maize to supra-optimal temperatures. Supra-optimal temperatures could
be more prevalent under future climate-change scenarios, but data sets in this study contained few data in that range.
A g ro n o my J o u r n a l Vo l u m e 10 6 , I s s u e 6 2 014
2087
T(1 h)
EnzymResp
Tmin, daily minimum temperature; Tmax , daily maximum temperature; T(1 h), estimated from daily Tmin and Tmax using sine-wave temperature distribution during a 24-h period (e.g., Tollenaar et al., 1979); Tmean, daily mean temperature;
T(3 h), estimated from range fractions that describe a sine wave.
T < 6C then T = 6C
T > 44.8C then T = 44.8C
Tmin and Tmax
TLU
MAIZSIM
Tmean
Tmean
GTIveg
GTIrep
CHU
T 8C
Reference
Daily rate
C d1
IR
(Tmin + Tmax)/2 10C
Thermal accumulation
Instantaneous rate (IR)
Boundary conditions
Input temperatures
(T)
Model name
Model type
Table 1. Summary of thermal functions, input temperatures (daily), temperature boundary conditions, and instantaneous and daily rates of thermal accumulation. The general thermal index (GTI) model has
separate functions for the pre-and post-flowering periods (GTIveg and GTIrep, respectively).
2088
2089
T -T T opt
R (T ) = Rmax ceil
Tceil -Topt Topt
/(Tceil -Topt )
[1]
where R(T) is the rate of development as a function of temperature. Yan and Hunt (1999) showed that this simplified b
function was highly predictive when model parameters that
were estimated from six constant temperatures were used
to predict the rates of 16 varying daynight temperature regimes in the Tollenaar et al. (1979) data set. The MAIZSIM
model (Kim et al., 2012) uses Eq. [1] for predicting the leaf
appearance rate and assigned the following parameter values:
Topt = 31.2C, Tceil = 43.7C, and Rmax = 0.53 d1.
6. Enzymatic response (EnzymResp): Using an equation for
describing the temperature response of enzyme activities
( Johnson et al., 1942), Parent et al. (2010) showed that
processes linked to plant growth and development showed
coordinated temperature responses and followed common
laws within a plant species. Interestingly, they showed that
there was no coordination of temperature responses linked
to plant metabolism and leaf photosynthesis. In a meta2090
R(T ) =
AT exp -DH A RT
a(1-T T0 )
1 + exp -DH A RT
[2]
2091
run for two different starting values, and the results were identical
in the two cases. This exercise confirmed that the algorithm was
converging to a global rather than to a local optimum.
In addition, for each of the five RM class in the 43-location
North America data set, the hybrid-location-years of the data
set were subdivided equally into three groups representing 33%
low, 33% medium, and 33% high means of daily minimum
temperature and of daily maximum temperature, resulting
in 15 mean-minimum and 15 mean-maximum temperature
groups (i.e., 5 RM classes 3 groups within each RM class).
This analysis is informative of thermal function sensitivity
variation by temperature. The thermal functions tested
vary in T base, Topt, and Tmax values. Differences in cardinal
temperature may render some thermal functions more sensitive
to specific temperatures than others, thus impacting their
relative precision in simulating development.
RESULTS AND DISCUSSION
Impact of Data Source on Thermal
Function Evaluation
Data source is an important variable when evaluating the
precision of thermal functions. The data sets evaluated in this
study each had limitations. The IndianaOhio data set was
comprised of a multilocation, multiyear study with orthogonal
treatments, but contained only a relatively small number of
hybrids and environments (i.e., location-years). In contrast, the
North American data sets consisted of a large number of hybrids
and environments, but the treatments were not orthogonal.
Different hybrids were grown in each location-year, resulting
in a different number of observations for each RM class in the
43-location data set (Tables 2 and 3). In addition, the silking
to black layer interval was evaluated only at a single geographic
location near DeKalb, IL, for multiple years (Table 3). The
Argentinean data set was not orthogonal and contained a
relatively small number of environments (years).
Empirical nonlinear functions were more precise than either
empirical linear or process-based functions in the North American
and IndianaOhio data sets, but the precision of the thermal
functions did not differ in the Argentinean data set (Tables 26).
Differences among thermal functions and calendar days were
generally smaller in the IndianaOhio than the North American
data set. The precision of the thermal functions for the planting
to anthesis phase in the North American data set were calendar
days < empirical linear < process based < empirical nonlinear,
whereas in the IndianaOhio data set they were calendar days <
empirical linear = process based < empirical nonlinear. In the
Argentinean data set, calendar days were less precise than the
thermal functions, but the thermal functions could not be
differentiated. Therefore, large data sets have improved ability to
differentiate thermal function performance, even when they may
have limitations such as lack of orthogonality. Ideally, large and
orthogonal data sets would be preferred.
Relative Precision of Thermal Functions
Within the large data sets that allowed the greatest
differentiation among the thermal functions, the empirical
nonlinear functions proved to be more precise than the other
functions tested (Tables 26). Thermal functions were generally
more precise than calendar days, although differences were
Agronomy Journal Volume 106, Issue 6 2014
Table 2. Coefficients of variation of the number of days and thermal accumulation during the planting to anthesis interval determined by eight functions for commercial hybrids across five relative maturity (RM) groups grown at 43 locations in the Corn Belt from 2007 to 2011 (North American
data set). The number of observations (n) is the sum of all hybrids grown in each location-year for a RM class.
Coefficient of variation
RM
7685
8695
96105
106115
116122
Mean
All
n
121
204
342
560
148
1375
Days
GDD10,30
CERES
MAIZSIM
TLU
EnzymResp
APSIM
GTI
CHU
%
10.3
9.5
9.3
9.0
8.4
9.0
7.8
7.6
7.0
9.9
9.7
9.5
9.2
8.5
9.2
7.7
7.6
6.8
9.3
6.4
6.5
6.2
6.0
6.3
6.0
5.7
5.6
10.4
6.4
6.9
6.0
5.3
6.2
5.4
4.7
4.5
10.3
6.3
6.9
5.9
5.3
6.0
5.2
4.5
4.6
10.1 e
7.7 d
7.8 d
7.3 cd
6.7 c
7.3 d
6.4 bc
6.0 ab
5.7 a
10.7
10.1
10.2
9.5
8.7
9.6
7.9
7.5
6.7
Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.
Table 3. Coefficients of variation of the number of days and thermal accumulation during the silking to black layer interval determined by eight thermal
functions for commercial hybrids across five relative maturity (RM) classes grown at a single location in the U.S. Corn Belt from 2007 to 2012 (North
American data set). The number of observations (n) is the sum of all hybrids grown in each location-year for a RM class.
Coefficient of variation
RM
7685
8695
96105
106115
116122
Mean
All
n
182
413
771
1477
286
3129
Days
GDD10,30
CERES
MAIZSIM
TLU
EnzymResp
APSIM
GTI
CHU
%
9.0
8.2
8.0
7.6
7.2
8.2
7.1
6.2
6.0
9.4
9.3
9.0
8.6
8.2
9.3
8.2
7.4
6.8
8.5
9.4
9.3
8.7
8.0
9.5
8.3
7.1
6.0
9.9
9.6
9.4
8.9
8.2
9.6
8.1
7.1
5.8
9.6
11.8
11.5
11.0
10.3
11.3
9.6
7.7
7.4
9.3 de
9.7 e
9.5 de
9.0 d
8.4 c
9.6 e
8.3 c
7.1 b
6.4 a
12.4
10.2
10.1
9.7
9.1
10.3
9.4
9.0
7.7
Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.
Table 4. Coefficients of variation of the number of days and thermal accumulation during the planting to silking and silking to black layer intervals determined by eight thermal functions for three maize hybrids planted at three dates from April to June during 1990 to 1994 at four locations in Ohio
and Indiana (Nielsen et al., 2002); the total number of observations is 108. Data were analyzed both separately for the early, medium, and late planting
dates and together for all planting dates.
Coefficient of variation
Planting date
Days
Early
Medium
Late
Mean
All
11.6
9.1
5.6
8.8e
12.9
Early
Medium
Late
Mean
All
9.5
11.5
8.6
9.9bc
10.3
GDD10,30
CERES
MAIZSIM
TLU
EnzymResp
APSIM
GTI
CHU
%
Planting to silking
4.4
5.2
4.7
3.6
4.5
5.9
4.4
3.5
4.7
5.4
4.9
3.9
4.9
6.1
4.5
3.8
5.7
6.4
6.1
5.2
6.1
7.2
5.7
4.7
4.9abc
5.7cd
5.3bcd
4.2ab
5.2bcd
6.4d
4.8abc
4.0a
5.2
5.8
5.5
4.7
5.5
7.0
5.4
5.0
Silking to black layer
8.4
9.2
9.0
7.9
8.7
10.1
6.8
8.0
9.0
10.2
10.0
8.6
9.4
11.0
7.6
8.9
11.3
12.7
12.5
10.7
11.8
14.0
8.6
10.5
9.6bc
10.7cd
10.5cd
9.0b
10.0bc
11.7d
7.7a
9.1b
11.8
13.1
12.8
11.0
12.0
12.8
7.9
10.5
Means within row followed by the same letter are not significantly different at 0.05 level of probability.
smaller during the period between silking and black layer than
during the planting to anthesis interval. The precision of the
thermal functions for both the pre-flowering and post-flowering
period in the North American data set was empirical linear <
process based < empirical nonlinear (Table 6). The CHU and
GTI functions were the most precise for the prediction of both
the pre-anthesis and post-silking phases in the North American
data set (Tables 2 and 3). In the IndianaOhio data set, the GTI
function was the only thermal function that was more precise
than calendar days for the silking to black layer period (Table 4).
The post-flowering GTI thermal function had a very different
2093
Table 5. Coefficients of variation of number of days and thermal accumulation by eight functions during the planting-silking and silking-black layer intervals of maize hybrids planted from September to January between 1989 and 2012 at locations near Balcarce (Argentina). Data were analyzed both
separately for four planting-date periods and together for all planting dates. The number of observations (n) is the sum of all hybrids grown in each
location-year for a RM class.
Coefficient of variation
Planting date
Sept.
Oct.
Nov.
Dec.Jan.
Mean
All
Sept.
Oct.
Nov.
Dec.Jan.
Mean
All
6
41
13
16
76
6
41
13
16
76
Days
GDD10,30
CERES
MAIZSIM
TLU
EnzymResp
APSIM
GTI
CHU
%
Planting to silking
5.4
8.2
7.8
7.7
7.6
7.6
6.7
7.0
6.9
6.7
6.2
6.5
6.2
5.6
6.4
6.5
6.3
5.5
9.0
8.5
8.7
8.5
8.3
8.7
8.6
8.4
8.1
12.5
9.9
10.0
9.9
10.0
9.9
10.2
10.2
10.8
8.4 a
8.2 a
8.2 a
8.1 a
7.9 a
8.2 a
8.0 a
8.0 a
7.8 a
18.5
8.8
9.1
9.0
9.0
9.2
10.3
10.1
10.7
Silking to black layer
6.7
6.0
5.0
5.4
5.7
5.6
5.1
5.4
5.4
12.5
12.6
13.0
12.7
12.1
12.8
12.9
12.7
11.9
18.7
11.0
10.3
10.8
11.9
11.0
11.8
13.6
13.3
11.0
8.6
8.5
8.4
8.4
8.1
7.6
8.2
8.7
12.2 b
9.6 a
9.2 a
9.3 a
9.5 a
9.4 a
9.4 a
10.0 a
9.9 a
13.0
17.0
17.5
16.8
15.7
17.0
15.7
14.1
14.0
Means within row followed by the same letter are not significantly different at 0.05 level of probability.
Table 6. Mean coefficients of variation of calendar days and thermal accumulation determined by empirical-linear functions (GDD10,30 and CERESMaize), process-based functions (APSIM, MAIZSIM, EnzymResp, and TLU), and empirical-nonlinear functions (GTI and CHU) during the planting to
anthesis or silking and the silking to black layer intervals in the 43-location, 5-yr and the one-location, 6-yr North American data sets (Tables 2 and 3),
the IndianaOhio data set (Table 4), and the Argentinean data set (Table 5).
Phase of development
Planting to anthesis or silking
Data set
North America
IndianaOhio
Argentina
North America
IndianaOhio
Argentina
Coefficient of variation
Thermal function
Calendar days
Empirical linear
Process-based
Empirical nonlinear
%
10.1 d
7.7 c
6.9 b
5.9 a
8.8 c
5.3 b
5.3 b
4.4 a
8.4
8.2
8.0
7.9
9.3 c
9.6 c
8.8 b
6.8 a
9.9 ab
10.1 b
10.3 b
8.4 a
12.2 b
9.4 a
9.4 a
9.8 a
Means within a row followed by the same letter are not significantly different at the 0.05 level of probability.
Table 7. Mean coefficient of variations (CV) of thermal accumulation determined by eight functions during the planting to anthesis and silking to black
layer intervals, and mean daily minimum and maximum temperatures of locations for five relative maturity (RM) groups. The planting-to-anthesis
interval comprised observations across 43 Corn-Belt locations during 2007 to 2011 (Table 1) and the silking to black layer interval comprised > 3100
observations at one Corn-Belt location from 2007 to 2012 (North American data sets).
RM
CV
7685
8695
96105
106115
116122
8.5 c
8.5 c
6.1 b
5.7 a
5.6 a
Planting to anthesis
Daily temperature
Min.
Max.
CV
%
C
11.7
23.1
12.1
23.9
13.0
25.4
14.3
26.3
15.7
27.5
C
17.1
28.4
16.9
27.9
16.6
27.7
15.6
26.7
15.1
26.3
7.3 a
8.4 b
8.3 b
8.3 b
10.1 c
Means within a column followed by the same letter are not significantly different at the 0.05 level of probability.
RM
d
7685
8695
96105
106115
116122
MAIZSIM
%
6.2 (23.0, 32.4)
9.0 (32.1, 43.7)
6.3 (25.7, 39.6)
9.2 (32.1, 43.7)
5.4 (26.2, 35.9)
6.2 (32.1, 43.7)
4.3 (27.6, 37.8)
6.0 (32.1, 43.7)
4.3 (30.3, 44.9)
5.9 (32.1, 43.7)
2095
2097